Android phones have gotten pretty fast over the last few years, to the point where faster processors don’t really bring about the noticeable speed improvements they used to. So now, the goalposts have shifted over to software — sometimes in places people don’t directly interact with.

Case in point: the latest improvement arrives in the form of something called Automatic Feedback-Directed Optimization (AutoFDO). AutoFDO is a kernel-level optimization, which is a pretty big deal, considering the kernel is the deepest layer of the operating system. It handles how your phone’s processor talks to apps, hardware, and everything in between. The last time we saw a performance enhancement that went this deep was perhaps the 2018 switch to a modern code translator for the Pixel 3.

As for what it actually does, the simplest explanation is that AutoFDO uses real-world data to make heavily used code faster. Android is written in code that your phone’s processor can’t directly understand, so a tool called the compiler has to translate it into something the hardware can actually run. Doing this can be pretty resource-intensive, though, so the compiler takes shortcuts. It makes tiny guesses, thousands of them, about how to optimize the code along the way. This is done by picking which functions matter most, or which paths are likely to be taken.

The problem is that these decisions are based on general rules and not on how people actually use their phones. AutoFDO changes this up by feeding the compiler real usage data. While there are a lot of great Android apps people don’t know about, this data was actually collected from lab tests where Google ran the 100 most popular apps on Pixel devices.

It’s worth mentioning that Google actually first introduced AutoFDO back with Android 12. But back then, it targeted apps and system libraries. Now, it goes deeper by directly targeting the kernel. In theory, this should bring about more significant gains since any improvements can ripple outwards. The kernel also eats up roughly 40% of your phone’s processing time, so any improvements here should be significant.

That said, while it all sounds pretty great, the actual results on Pixel hardware were modest. Cold app launches got 4% faster, and boot times dropped by around 2%. But certain internal benchmarks, like Binder tests, did improve by a far more notable 21%. Google also claims these changes should improve your Android’s battery life, since the processor ends up doing less redundant work.

Google has also built a whole pipeline that refreshes the data profiles feeding into AutoFDO continuously. What this means is that as Android’s code evolves over time, the optimizations stay relevant instead of going stale.

Looking further out, Google wants to push AutoFDO beyond just the main kernel binary. The team has talked about extending it to additional kernel modules. They’ve also mentioned adding it to the custom hardware drivers that phone manufacturers add for things like cameras and modems. Right now, kernel AutoFDO is live on the Android 16-6.12 and Android 15-6.6 kernel branches, with plans to expand into Android 17-6.18 soon. Of course, none of this matters if manufacturers don’t update your Android phone’s kernel to include the new profiles.